agglomeration$shadow:$!4!/39!!...

1 / 39

Agglomeration Shadow: Can Investment and Education Alleviate Geographic Disadvantage in Urban Growth?

Zhao Chen, Ming Lu and Zheng Xu∗

Abstract: Using Chinese city-‐level data from 1990 to 2006, this paper provides evidence for the

non-‐linear CP model of urban systems (Fujita and Krugman, 1995; Fujita, Krugman and Mori

(1999). Our results show that a proximity to major ports and international markets is essential

for urban growth. Moreover, the geography–growth relationship follows the ∽-‐shaped nonlinear

pattern, following which Middle China lies in the agglomeration shadow of Eastern China. We

also find that in the long run, geography plays a much more important role than in the short run.

Investment and government expenditure can increase short-‐run growth, but not significantly in

the long run. Only educational investment can help alleviate geographic disadvantage for laggard

areas in the long run.

Key Words: core–periphery model, urban systems, investment, education, China

∗ Zhao Chen: China Center for Economic Studies, Fudan University, Shanghai, 200433, China, Email: [email protected]; Ming Lu: Corresponding author, Department of Economics, Shanghai Jiaotong University and Fudan University, Email: [email protected]; Zheng Xu: University of Connecticut, Storrs, CT 06269, U.S. E-‐mail: [email protected]. Financial support from the National Natural Science Foundation (71133004, 71273055), and the National Social Science Foundation (12AZD045) are greatly appreciated. This research is also supported by the Shanghai Leading Academic Discipline Project (B101) and Fudan Lab for China Development Studies. We thank Jacques-‐François Thisse, Stephen L. Ross, Thierry Mayer, Dao-‐Zhi Zeng, Wing-‐Thye Woo, Robert Feenstra, Deborah Swenson, Chun Chung Au and seminar/conference participants at ADB conference in Hawaii, UC Davis, Bank of Finland, Fudan University, Peking University, Chinese University of Hongkong, and Hongkong University of Science and Technology for their useful comments. The content of this article is the sole responsibility of the authors.

2 / 39

I. Introduction

For countries with vast territory, interregional balanced development is an essential policy

goal. However, in a modern economy, industrial agglomeration and uneven distribution of

production lead to interregional inequality, especially when labor is not perfectly mobile. In the

globalization era, coastal regions near ports are more advantageous to develop international

trade and attract FDI than interior regions are. To counteract the agglomeration forces of trade,

central governments usually pursue interregional balanced development by central-‐to-‐local fiscal

transfer and government-‐sponsored projects. However, some regions stay poorer persistently

than others. Therefore, how important geography is in shaping interregional growth pattern and

what kind of policies can help laggard regions grow faster remain questions to be explored.

From its position as the backbone of new economic geography (NEG) theory (Krugman,

1991), the core–periphery (CP) model suggests the existence of an uneven distribution of

economic activities in spatial economy caused by the interplay of scale economies and transport

costs. As proved by Fujita and Krugman (1995), Fujita, Krugman and Mori (1999), a nonlinear

urban system indicates a ∽-‐shaped correlation between the distance to the core and local market

potential. In this urban system, regions close to the core are under the “agglomeration shadow”,

so their economic resources are absorbed by the core. However, current empirical studies using

US and European data barely test the presence of such nonlinearities (Black and Henderson,

1999; Dobkins and Ioannides, 2004; Brülhart and Koenig, 2006). As Krugman (1991) points out,

the share of manufacturing in national income matters in the presence of the CP pattern. As a

consequence, the decreasing share of manufacturing in both the US and Europe may give some

indication as to why previous studies have not filled the gap between theory and reality.

Using Chinese city-‐level data from 1990 to 2006, this paper proves the nonlinear CP pattern

in Chinese urban systems by estimating the impact of the distance to major ports on urban

economic growth. Industrialization and globalization have been two important forces

underpinning rapid economic growth in China since the 1980s. Being a rapidly industrializing

country with a vast geographic territory, China provides an appropriate context to examine how

geography influences the spatial distribution of economic activities.

More particularly, when joined with globalization and developments in export-‐led

manufacturing, coastal ports and nearby cities have greater access to international markets, thus

providing key advantages for economic growth (Hanson, 1998). Therefore, in this paper, we

assume that there are two national-‐level hierarchical monocentric urban systems in China, the

cores of which are the major ports and megacities of Shanghai and Hong Kong. Further, as

predicted by the CP model, we hypothesize that the spatial interaction within each urban system

follows a ∽-‐shaped nonlinear relationship between the distance to the port and urban growth,

and test this using empirical methods. Our key finding is that a ∽-‐shaped nonlinear correlation

indeed exists in relation between the geographic distance of cities to major ports and economic

growth in China’s urban systems. Consequently, the economic forces entailed by the CP pattern

are currently reshaping Chinese economic geography.

3 / 39

The empirical CP model also enables us to compare the impacts of geography and other

growth factors. We find that the distances to major ports and regional agglomeration centers are

much more powerful than other variables. In the long run growth over the time period of

1990-‐2006, only the two distance variables explain 6.9% of the growth variances, while a full model

comprising all the growth factors can explain 40.4% of the growth. In contrast, investment-to-GDP

ratio and government expenditure-to-GDP ratio are growth-enhancing factors in the short-run growth,

but they are insignificant in the long run. Only education investment measured by teacher-student ratio

can boost long-run growth and help periphery regions alleviate their geographic disadvantage.

The remainder of the paper is structured as follows. Section II briefly reviews the theoretical

and empirical work on spatial interaction between cities. Section III introduces details of China’s

urban system and explains why it represents a feasible application of the CP theory. Section IV

discusses the data and our chosen econometric approach. Section V presents the basic results.

Section VI extends the model to medium-‐ and long-‐run growth regressions, and compares the

relative importance of geography and other variables. Section VII provides some concluding

remarks.

II. Literature Review

The NEG theory emphasizes the interplay of centripetal and centrifugal forces in

determining spatial economies (Krugman and Elizondo, 1996). Centripetal forces here include

both pure external economies and a variety of market scale effects, such as forward and

backward linkages, while centrifugal forces comprise pure external diseconomies, such as

transport costs and land rents (Krugman, 1991). For the most part, the CP model formalizes the

role of agglomeration and dispersion in the dynamic formation of a monocentric urban system, of

which a prominent feature is the emergence of a hierarchy of cities based on market potential,

featuring especially a symbiotic relationship among cities (Fujita, Krugman, and Mori, 1999).

Fujita and Krugman (1995), Fujita and Mori (1997), and Fujita, Krugman, and Mori (1999)

assume that manufacturing concentrates in the core of these systems because of the presence of

scale economies, while the periphery is the agricultural hinterland. Near the core, the market

potential curve decreases by distance as the ever-‐closer proximity to demand and supply brings

about scale economies and lowers transportation costs (Venables, 1996). However, as

transportation costs increase by distance, firms producing goods that are close substitutes may

find that they will earn more if moving away from the core city and instead serve customers in

fringe areas (Fujita and Krugman, 1995).

In this case, the market potential curve will increase at distances further away from the core.

However, as distance increases, the potential curve will eventually fall again for similar reasons

as it did nearer the core. Using numerical simulation, the CP pattern is then a ∽-‐shaped

relationship between the distance to the core and market potential in the periphery in a

monocentric urban system. This curve shows that as the distance to the core increases, the

market potential in the hinterland first declines, when the agglomeration effects are stronger and

4 / 39

later rises as dispersion forces dominate. Later, the curve will again decline for more remote

areas. Consequently, an agglomeration shadow exists for places not far enough away from the

core (Fujita and Krugman, 1995; Fujita and Mori, 1997; Fujita, Krugman, and Mori, 1999).

As Fujita and Krugman (2004) conclude, “…buttressing the approach with empirical work” is

one of the most important directions for future research in NEG. Based on the implications of CP

theory, Dobkins and Ioannides (2001) suggest that nonlinearities should appear in the spatial

interactions in urban systems. Consequently, directly examining the spatial interactions among

cities in relation to the geographic distance between them, their urban economy, or their place in

the urban hierarchy, has generally become one of the streams of empirical research, not only

concerning agglomeration (Hanson, 2001), but also the spatial distribution of cities (Partridge et

al., 2010). However, “…due to the highly nonlinear nature of geographical phenomena”, it’s not

easy “…to make the models consistent with the data” (Fujita and Krugman, 2004).

For the most part, the empirical work on spatial interaction has partly verified the CP model

and the role of agglomeration forces in urban systems, more or less finding that closer (to core) is

better. Nonetheless, the nonlinearity of the CP model remains unverified. For instance, Combes

and Overman (2004) use a geographic information system map to depict gross domestic product

(GDP) per capita at the country level in Europe in 1996, thereby suggesting the spatial

distribution of economic activity in the European Union follows the CP pattern. However, there is

no evidence of nonlinearity. More to the point, although maps of wages and employment can

display similar patterns, they also suggest that this pattern may not as marked in terms of total

GDP. Using regional data over 1996–2000, Brülhart and Koenig (2006) analyze the internal

spatial wage structures of the Czech Republic, Hungary, Poland, Slovakia, and Slovenia, and find

that real wages fall with distance to the national capital and Brussels. However, none of the

estimates is statistically significant.

The empirical evidence drawn from the US urban system is even more complex and

puzzling. For example, Black and Henderson (1999) examine the correlation between population

growth in US metropolitan areas and initial conditions over the period 1950–1990. They

conclude that population growth is faster in cities closer to the coast and to cities with larger

initial populations; though this effect weakens as neighboring population masses become larger.

Similarly, Dobkins and Ioannides (2001) and Ioannides and Overman (2004) employ US

metropolitan data over the period 1900–1990 and conclude that distance from the nearest

higher-‐tier city is not always a significant determinant of urban size and growth. Moreover, there

is no evidence of the nonlinear effects of either size or distance on urban growth over such a long

sample period. Finally, Partridge et al. (2010) explore whether the proximity to higher-‐tiered

urban centers affected US county population growth over the period 1990–2000. Their results

suggest that larger urban centers indeed promote growth in more proximate places, but only

those with fewer than 250,000 persons. As a result, Partridge et al. (2010) conclude that NEG

theory (and the CP model) only partly explains modern US urban growth, thereby suggesting the

need for a broader framework.

5 / 39

Encouraged by the above studies, we believe the time is right to clarify whether the CP

model is relevant in explaining contemporary urban systems. We also believe that an empirical

study of the Chinese urban system is most suitable for testing nonlinearities in the CP model,

especially as unlike the US and Europe, China is still experiencing urbanization and

industrialization. During this critical process, transport costs and scale economies are essential

determinants of the spatial distribution of urban economies. Besides, the globalization process is

also helping to reshape China’s economic geography and urban system, making coastal areas

advantageous in attracting foreign direct investment (FDI) and developing export-‐led

manufacturing (Wan, Lu, and Chen, 2007). Whether the evolution of the urban system in China

approximates the purported nonlinear CP pattern is then an interesting topic and a useful

empirical exercise. If China’s urban growth indeed follows a nonlinear CP pattern, it is relevant to

explore how important geography is relative to other growth factors and whether investment

and education can alleviate geographic disadvantage of the cities in the agglomeration shadow.

III. China’s Urban System Evolution and an International Comparison

Krugman (1991) questions, “…how far will the tendency toward geographical concentration

proceed?” Taking early nineteenth-‐century America as an example, with a small share of

manufacturing employees, a combination of weak economies of scale, and high transportation

costs, Krugman (1991) believes that pre-‐industrial society was satisfied by small towns serving

local market areas. Consequently, the answer to the question posed earlier depends on the

underlying parameters of the economy, such as the share of manufacturing, the presence of

economies of scale, and transportation costs. However, Fujita and Krugman (1995) find that a

reduction in transportation costs is most likely to cause a population migration from the city to

the hinterland. Therefore, with the dramatic development of transportation technology and

declining manufacturing shares, postindustrial societies, like the US and Europe, have since the

late 1970s begun to exhibit a trend in “anti-‐urbanization” (Frey and Speare, 1992).

When compared with the US or Europe (e.g. France), China is still in the process of

industrialization. According to the World Bank (2011), in 2008 the share of industrial

value-‐added in GDP1 in China was 47.4%, significantly more than in the US (21.3%) or France

(20.3%), both of which were continuing to decrease year by year, as shown in Figure 1.

Consequently, the large share of manufacturing in the Chinese economy combined with rapid

economic growth makes China a better trial to test the CP model of urban systems. Besides, with

the force of spatial agglomeration brought about by manufacturing, the rapid development of

international trade continues to reconstruct China’s urban economies, especially with China’s

increased openness to the world starting in the 1990s. This opening process is also a natural

experiment to consider the role of distance to major ports and access to international markets in

changing the urban system.

1 Industry value-‐added corresponds to ISIC divisions 10–45, including manufacturing (ISIC divisions 15–37). It therefore includes value added in the mining, manufacturing, construction, and electricity, water, and gas industries.

6 / 39

[Insert Figure 1 here]

3.1. Economic Openness and Urban System Adjustment

International trade changes the reference market for firms in a country by shifting resources

to locations with low-‐cost access to foreign markets, such as border regions and port cities. Fujita

and Mori (1996) propose a model of spatial economic development in which agglomeration

economies and the hub effect of transport nodes interplay. Their simulation of a monocentric

urban system shows a nonlinear relationship between market potential and distance to the core

city. In this model, if the core city is a port, openness and international trade will strengthen the

agglomeration effects of the core city and reshape the urban system within a country.

Corresponding to the theoretical implications, Hanson (1998) finds that following trade reform

and increased global openness, especially with the US, firms in Mexico began to locate in regions

with good access to foreign markets, such as Mexican cities bordering the US.

A similar process of economic reform and openness began in China in 1978. After 1992,

China quickly opened its doors to the world market, with reforms invoking a widespread opening

up of markets and a commitment to the market economy (Ho and Li, 2010). Figure 2 depicts the

increasing importance of international trade in China, especially after 1992. In 1994, the official

and black market exchange rate for Renminbi (RMB) combined. Since then, China has exhibited

an export-‐led growth pattern and has continuously increased its trade surplus to the present day.


Besides the dramatic development in international trade and the rapid and steady GDP

growth, another remarkable phenomenon in China is the rate of urbanization, increasing from

17.9% in 1978 to 49.68% in 2010,2 almost one percentage point per year. What exactly is the

linkage between international trade and urban system adjustment? Again taking Mexico as an

example, Krugman and Elizondo (1996) find that the trade policies of developing countries and

their tendency to develop huge metropolitan centers closely link. Fujita, Krugman, and Mori

(1999) also argue that rapid urbanization in the world economy infers the increasing importance

of cities as the basic unit of international trade. Therefore, international trade may not only

reshape regional and national economic geography, but also greatly affect the evolution of the

urban system.

Ever since the mid-‐1990s, with China increasingly open to the international market,

international trade has become more and more important for Chinese urban development. For

instance, Wei (1995) finds that Chinese cities with large export sectors generally grow faster.

Further, international trade strengthens the agglomeration force in major ports in China. For

example, Wei (1995) and Anderson and Ge (2005) provide some evidence that many emerging

2 Source: NBS, The Statistical Summary of the 6th Population Census, http://www.stats.gov.cn/tjfx/jdfx/t20110428_402722253.htm.

7 / 39

cities locate in coastal areas in the Pearl River Delta near Hong Kong. More and more empirical

studies are finding that recent economic agglomeration has changed the distribution of China’s

regional and urban economies, thus coastal and large cities have higher economic growth rates

(Chen et al., 2008; Bao et al., 2002).

For China’s cities, which had hitherto developed evenly under central government planning

before economic reform, the open-‐door policy represented an exogenous shock to city growth

and the urban system. Unlike the 1980s, when China selected several coastal cities for opening, in

the 1990s, the national policy was an open door. The geographic difference between cities,

mainly the distances to major ports in determining international trade costs, plays an essential

role in international trade, FDI inflow, and city growth. Therefore, the opening process in the

1990s is a natural experiment for economists to see whether and how the urban system evolves

following a CP model.

3.2. Geography and Growth: Why China Can Contribute to NEG Studies?

Partridge et al. (2010) believe agglomeration economies may have a greater geographic

scope than usually assumed by economists. From this perspective, economic geography,

particularly the role of nonlinear spatial agglomeration, may require wide spaces and a long

timeframe for accumulation, while national and geographic boundaries, wars, and other factors

often limit the free flow of resources. As a result, it is sometimes difficult to use econometric

models and real-‐world data to portray how agglomeration actually works. Therefore, studies of

stably developing large countries, such as the US (and China after 1978) are particularly

important for empirical research on NEG theory.

Research using Chinese data may helpfully contribute to the empirical study of NEG for the

following reasons: First, China has a vast territory, as well as a large population. The vast

territory not only provides the space required by agglomeration, but also plenty of samples for

empirical study. Meanwhile, a large population provides sufficient market potential. Second,

compared with the US, China has a larger interregional geographical diversity and a more

obvious geographical heterogeneity because of the concentration of ports along the eastern coast.

Interregional geographic differences are significantly less in the US, which has large ports on both

its western and eastern coasts. Finally, China has witnessed rapid economic development in the

past thirty years, especially after 1992, with corresponding changes in the spatial distribution of

economic activities. This allows us to better observe the impact of spatial agglomeration on the

urban system.

3.3. China’s Urban System

In China, a “city” is a local administrative and jurisdictional entity, and there are comparable

historical data and large enough cross-‐sectional observations of cities. There are three different

administrative levels of cities in the Chinese urban system: municipalities, prefecture-‐level cities,

and county-‐level cities. This does not include smaller settlements with townships or lower

8 / 39

administrative levels. The major administrative criteria announced in 1963 defined a city as an

urban agglomeration with a total urban population of greater than 100,000 inhabitants. The

economic and political importance of urban agglomeration is also one of the government’s

considerations in defining cities. However, the definition of cities has generally been consistent

since at least 1949 when the People’s Republic of China was established (Anderson and Ge,

2005).

China’s National Bureau of Statistics (NBS) reports information about both prefecture-‐ and

county-‐level cities, but we use only data on prefecture-‐level and above cities in our research. As

for county-‐level cities, their boundaries and number have fluctuated greatly in recent history (see

Table 1). The number of prefecture-‐level cities has also increased greatly since 1990, so the

boundaries of the prefecture-‐level cities may also have changed. However, for cities at the

prefecture level and above, NBS reports information on both Diqu (urban area plus rural counties

within the same administration) and Shiqu (urban area only). In general, Shiqu is a good

definition of the metropolitan area by international standards (Fujita et al., 2004), and the

entities implied by new cities usually arise outside the Shiqu, so this information is comparable

across time and is thus used in our study.

[Insert Table 1 here]

The definition of China’s urban hierarchy is another problem. As mentioned above, three

administrative levels of cities are included in China’s urban system: municipalities (or

province-‐level cities), and prefecture-‐ and county-‐level cities; provincial capitals also comprise a

special administrative level in prefecture-‐level cities. Cities of different levels have different

population sizes. However, corresponding to the CP model, our study focuses on the spatial

agglomeration in urban system, for which urban size (e.g. population) and market potential are

much more important than administrative level. Accordingly, we distinguish between cities using

different definitions and sizes, as shown in Table 2. In the model specification, we use dummies

to control for administrative level, special economic zones (SEZ), and city size, etc.


IV. Model Specification and Data

4.1. Model Specification

Our model is used to estimate how geographical location, mainly distance to a major port,

affects urban growth. The reduced-‐form estimation, where geography is the focus, is widely used

in earlier empirical studies of the spatial interaction between cities (Brülhart and Koenig, 2006;

Dobkins and Ioannides, 2000). We also use cross-‐sectional ordinary least squares (OLS)

regression for our basic reduced-‐form CP model of the urban system, as based on the economic

growth model in Barro (2000). As the urban system is constantly evolving during the opening

process, the different growth rates of cities in different locations will reflect the shifts in the

9 / 39

urban system. In addition, given all the explanatory variables are exogenous geographic factors

or the initial values of the control variables, the potential endogeneity problem in OLS estimation

is not a major concern. The model specification is as follows:

Dgdpit = f (lngdpi0, invei0, labi0,edui0,govi0, fdii0,coni0,geoi0,)

Per capita GDP growth is the dependent variable. The extant literature also uses urban per capita

GDP growth as a measure of the urban economic activities (Glaeser et al., 1995; Dobkins and

Ioannides, 2000). We do not employ real wages, income, population, and per capita GDP in our

analysis for the following reasons. First, wage and income data are not available at the city level

in China. Second, in the early years of our sample, the urban population in China did not include

migrants without local household registration identity (hukou) (Au and Henderson, 2006b), so it

does not well represent the urban economy or size. Third, as we wish to illustrate the dynamics

of urban economies following China’s openness to the world, per capita GDP growth (not the

level of GDP), should be a better measurement to observe the evolution of the urban system.

Nevertheless, we also estimated models specifying per capita GDP instead of GDP growth as the

dependent variable. However, our main results concerning the Chinese urban system were

unchanged.3

In our study, we highlight the roles of openness and international trade in determining the

distribution of China’s urban economic activities, so the key variable of geography is the city’s

distance to the nearest major port. We construct our dataset over the period 1990–2006 to

capture best the drastic openness in place since the early 1990s. Problematically, most of the

information about China’s cities in 1992 and 1993 is missing from the Chinese Cities Statistical

Yearbook (National Bureau of Statistics, 1991–2007). Consequently, we use the data for

1994–2006 to check the robustness of our results after 1994 (coinciding with the period of

far-‐reaching openness and depreciation of the RMB).

Hanson (2001) suggested that three issues might cause trouble in identifying agglomeration

effects in empirical studies: (i) unobserved regional characteristics, (ii) simultaneity in regional

data, and (iii) multiple sources of externalities, of which the former two are particularly crucial

when it comes to the spatial interaction of cities. As for “unobserved regional characteristics,” we

attempt to control for as many variables as the data will allow according to both the literature on

economic growth and past empirical studies of China. “Simultaneity in regional data,” however,

may be a minor issue in our study, as the core explanatory variables in our model are geographic

and do not change. As for the other explanatory variables, we use their initial values in 1990 to

counter simultaneity bias in the model. As such, the estimated results represent the long-‐term

impact of these explanatory variables on urban economic growth. Besides, it is quite uncommon

to control for so many initial state variables in the empirical studies employing NEG theory.

Therefore, we also estimate several models including only geographic variables. However, the

basic model follows the economic growth literature to control for other factors influencing

growth and this helps control for heterogeneity across cities.

3 For brevity, we do not discuss these results, but they are available from the authors upon request.

10 / 39

4.2. Data

In this paper, we use China’s urban area (Shiqu) data (1990–2006) complied from the

Chinese Cities Statistical Yearbook (National Bureau of Statistics, 1991–2007), including 286 cities

at the prefecture-‐ and above level4 from 30 provinces across mainland China. Hong Kong, Macao,

Taiwan and Tibet are not included. Local statistical bureaus in China have for some years

collected data on all enterprises in their local area, and report GDP figures at the level of the

appropriately defined metropolitan area (Fujita et al., 2004; Au and Henderson, 2006a). Though

we may express some doubt as to the quality of statistical data in China, these GDP figures are of

a high quality, as discussed by Au and Henderson (2006a), especially when econometric models

find highly significant results that are consistent with established economic theory. The

remaining data sources and the construction of the variables are as follows.

4.2.1. Urban Economic Growth

The dependent variable Ddgpit is the compound average annual growth rate of real per

capita GDP from 1990–2006 deflated by the urban consumer price index (CPI) in each province

for city i over t years. The theoretical assumptions of NEG emphasize the agglomeration of

manufacturing and services (Krugman, 1991), so we correspondingly removed agricultural

output and population from the GDP and population indicators, respectively.

4.2.2. Spatial Interactions among Cities

Hanson (1998, 2005) approximates the access of each region to its principal markets using

geographic distance. Therefore, geographic distance, as well as driving or road/railway distance,

is used to approximate the spatial interaction among cities (Dobkins and Ioannides, 2000;

Brülhart and Koenig, 2006; Partridge et al., 2010). In particular, we use straight geographic

distances to major ports as our measure of spatial interaction, which avoids the potential

endogeneity bias brought about by measures of traffic distance, such as driving hours.

We assume that there are two hierarchical monocentric urban systems in China. The first is

the national urban system, the cores of which are the major ports, like Shanghai and Hong Kong.

The distances to these ports measure the remoteness of cities to the global market. The second is

the regional urban system(s), the cores of which are large cities like Guangzhou, Chongqing, and

Wuhan, etc. The spatial interaction within each urban system exhibits the CP structure, which we

testify to using our empirical results. Here, we use the distance to the nearest “big” city (distbig)

to measure the interaction within regional urban systems, and the distance to the nearer major

port (disport) to measure the interaction within national urban systems.

To exploit how the agglomeration of major ports affects urban economic activities in China’s

spatial system, we need to identify “major ports.” According to the China Statistical Abstract 2009,

in 2008, Hong Kong accounted for 10.8% of overall Chinese cargo throughput, Shanghai 10.6%,

4 We have data on 286 cities in 2006 and 211 cities in 1990.

11 / 39

and Tianjin 7.4%.5 As to ports in different regions, the Bohai Sea ports (Tianjin, Qingdao,

Qinhuangdao, Dalian) in total accounted for 24% of overall cargo throughput, with the Pearl

River Delta ports (Hong Kong, Guangzhou, Shenzhen) representing 22.4%, and the Yangtze River

Delta ports (Shanghai, Ningbo-‐Zhoushan) another 21.4%. The above data clearly show that cargo

throughput in China mainly concentrates in these three areas. However, the distribution of the

Bohai Sea ports is widely dispersed. Therefore, agglomeration in Tianjin may not be as strong as

in Shanghai and Hong Kong. Therefore, we initially define Shanghai and Hong Kong as “major

ports” in our study but include Tianjin when testing for robustness. In fact, Shanghai and Tianjin

are provincial-‐level cities and respectively major ports in the Yangtze River Delta and Bohai Belt

city cluster. Likewise, Hong Kong is the major port of the Pearl River Delta and an international

financial center. Hong Kong is also adjacent to Shenzhen, another economic center and major

port in the Pearl River Delta. Thus, the distance to Hong Kong actually represents the distance to

Hong Kong and Shenzhen.6 See Figure 3 for the location of the major ports.


In Table 2, we attempt to reveal a proper definition for higher-‐level cities in the regional

urban system using their nonagricultural population. To construct time-‐consistent data, we only

use the information on cities in 1990 to identify large cities, with the information after 1990

merely used to illustrate how quickly urban sizes in China have evolved. We identify a

nonagricultural population of 1.5 million as the lower boundary of large cities in China’s regional

urban hierarchy. The reasons for this criterion are as follows. First, in 1990, there were nine

cities with a nonagricultural population greater than two million, eight of which located in coastal

areas. Accordingly, this would not well represent the urban hierarchy in inland areas. Second, in

1990, there were 31 cities with a nonagricultural population of more than one million, most of

which were municipalities and provincial capitals. Therefore, high-‐level cities defined according

to this criterion may correlate highly with municipalities and provincial capitals in China.

Moreover, if we set a nonagricultural population of one million as the lower boundary, there are

simply too many large cities defined.

Considering these, we define large cities in China’s urban hierarchy as those with a

nonagricultural population of more than 1.5 million in 1990. Our assumption is that via

agglomeration, these 14 large cities either already had or would generally become the regional

centers of the domestic and regional market and thus play a central role in China’s urban system

for each region within the CP structure. Figure 3 depicts the distribution of large cities in China.7

5 Because of the different measurements of cargo throughput used, Shanghai sometimes ranks higher than Hong Kong. Fortunately, this is not a major issue in our study as both are classified as major ports. 6 We compile our information about port cargo throughput from the China Statistical Abstract 2009 (National Bureau of Statistics, 2009). Though this publication also contains information on ports in mainland China and Hong Kong in 1990, measurement differs and they are therefore difficult to compare. Therefore, we were obliged to use the most recent year (e.g. 2008) of data for cargo throughput using the same measurement (tons). 7 These are Shanghai, Beijing, Tianjin, Shenyang, Wuhan, Guangzhou, Ha’erbin, Chongqing, Xi’an, Nanjing,

12 / 39

As shown, almost all are provincial capitals, with the exception of Dalian. To test whether the

results are sensitive to our definition of large cities, we substitute the distance to the nearest

large city for the distance to the nearest municipality or provincial capital to check the

robustness of our empirical results (as discussed in Section VI). We measure geographic distance

using China Map 2008, issued by Beijing Turing Software Technology Co., Ltd., China Transport

Electronic & Audio-‐Video Publishing House. As the CP model displays nonlinear characteristics,

we also include the squared and cubed values of distance in some of our regressions to capture

any nonlinearity.

For the 14 large cities defined, their distances to the nearest large city are all zero. Their

economic growth may benefit from their city size, so we use a dummy largecity to control for this

effect. However, this should not be a problem for our major ports as Hong Kong is not included in

our sample and the largecity dummy already controls for Shanghai. In addition, the GDP level of

the nearest large city in the initial year is also controlled for by gdpofbig, which is ln(GDP) in

1990, exclusive of agricultural output.

4.2.3. Market Segmentation and Border Effect

Policy is an important factor affecting spatial agglomeration in NEG theory. When it comes to

studies on the Chinese urban system, an important policy factor that affects regional economic

development is interregional market segmentation. For example, several studies have suggested

that there is a serious market segmentation problem among Chinese provinces (Young, 2000;

Ponect, 2005; Chen, et al., 2007). Interprovincial division is likely to increase the actual distance

between cities in different provinces, and limit the spatial interaction among cities. Our study

attempts to reveal whether the administrative boundary limits spatial interaction within the

urban system. Denoted as samepro, the dummy variable represents whether a city is in the same

province as its nearest large city. If the “border effect” of China’s provinces does not affect spatial

interaction between cities (and thus urban economic growth) this variable should be

insignificant.

4.2.4. Other Geography Variables

In fact, spatial interaction is only the second nature of cities. For studies of China’s urban

economy, there are some other geographic factors of first nature that we should control for that

may, according to Krugman (1993), also indicate potential initial advantages.

Mainland China is usually divided into three regions: East, Center, and West. The Eastern

region has eight provinces and three municipalities, Central China has seven provinces, and the

West has eleven provinces and one municipality. There are significant interregional differences

in climate, access to water, topography, and other features of the environment (Ho and Li, 2010),

along with economic development. National-‐level policies are also designed for achieving

interregional economic balance between the East, Center, and West. Therefore, we use dummy

Dalian, Chengdu, Changchun, and Taiyuan.

13 / 39

variables for “west” and “center” to capture any interregional differences in geography and

policies.8 However, the regional division of mainland China does not strictly follow geographic

location. For example, Guangxi Province is geographically located in the south coastal area, but

grouped by the Strategy of Western Region Development launched by Chinese government in

1999 with the West given its poor economic performance. Figure 3 illustrates the distribution of

China’s provinces.

Cities with ports in coastal area or with river ports along large rivers may also benefit from

access to international or domestic markets, so we also specify seaport and riverport as dummies

to control for this initial geographic advantage. The list of cities with seaports or river ports is

from “The First China Port City Mayors (International) Summit 2006” held by the Development

and Research Center, the Ministry of Communications, the Municipal Government of Tianjin, and

the China Communications and Transportation Association.9

4.2.5. Initial State of Economic Growth Factors

Following the traditional economic growth literature (for example, Barro, 2000) and studies

on China’s economic growth, we will also measure some other factors that may contribute to

China’s urban economic growth, such as initial economic performance, investment, labor,

education, etc.

lngdpi0 is the logarithm of the per capita output of manufacturing and service in 1990, which

is controlled for to observe whether the Chinese economy experiences conditional convergence

at the city level.

invei0 is the ratio of investment to GDP, where GDP is the total output of manufacturing and

services in 1990.

We use the ratio of teachers to students in primary and junior schools (edui0) in 1990 to

control for the effect of education; this may not be a good measurement, but is the best we could

identify. We specify the ratio of employees to total population (labi0) in 1990 to proxy for labor.

China’s urban labor market reforms radically pushed ahead in the mid-‐1990s, so in 1990 the

number of urban employees contained a substantial amount of disguised unemployment in state-‐

and collective-‐owned enterprises. Therefore, a high labor ratio may not be significantly positive

for urban economic growth.

The government expenditure-‐to-‐GDP ratio and FDI-‐to-‐GDP ratio in 1990, as usually

controlled for in the economic growth literature, are respectively govi0 and fdii0, where GDP is the

8 West includes cities from 12 provinces and one municipality, namely, Chongqing, Sichuan, Yunnan, Tibet, Shaanxi, Gansu, Qinghai, Ningxia, Xinjiang, Inner Mongolia, Guangxi, and Guizhou. Center includes cities from seven provinces, namely, Hebei, Anhui, Jiangxi, Henan, Hubei, Hunan, and Shanxi. The cities of the remaining 12 provinces and municipalities are in the East. These are Heilongjiang, Jilin, Liaoning, Tianjin, Beijing, Shandong, Jiangsu, Shanghai, Zhejiang, Fujian, Guangdong, and Hainan. 9 There are 32 cities with seaports: Qingdao, Yantai, Weihai, Rizhao, Haikou, Sanya, Tianjin, Tangshan, Qinhuangdao, Cangzhou, Dalian, Jinzhou, Yinkou, Lianyungang, Fuzhou, Xiamen, Quanzhou, Zhangzhou, Guangzhou, Shenzhen, Zhuhai, Shantou, Zhanjiang, Zhongshan, Shanghai, Ningbo, Wenzhou, Zhoushan, Taizhou, Beihai, Fangchenggang, and Qinzhou. There are another 22 cities with river ports: Ha’erbin, Jiamusi, Wuhu, Ma’anshan, Tonglin, Anqinq, Yueyang, Nanjing, Wuxi, Suzhou, Nantong, Yangzhou, Zhenjiang, Foshan, Dongguan, Luzhou, Wuhan, Yichang, Nanchang, Jiujiang, Nanning, Wuzhou, and Chongqing.

14 / 39

total output of three sectors, as we cannot distinguish between the final flows of government

expenditure and foreign direct investment across sectors. The effects of government expenditure

and foreign direct investment on economic growth are conventionally difficult to predict

beforehand (Barro, 2000; Clarke, 1995; Partridge, 1997).

coni0 represents several other control variables related to Chinese urban economic growth,

including the ratio of the nonagricultural population to the total population (urb) (to account for

the level of urbanization) and population density (density) and its square (density2) (to reflect

internal population agglomeration in urban areas). We also use the ratio of service GDP to

manufacturing to proxy for the industrial structure of the urban sector (SM).

4.2.6. Policy

We can never ignore policy in empirical studies of agglomeration and urban economies.

Other than the administrative boundaries mentioned, there may be some other policy advantages

affecting China’s urban economic growth that are controlled for in the literature (Bao et al., 2002;

Ho and Li, 2010).

As discussed earlier, in our prefecture-‐level database, different administrative levels of cities

remain, namely, municipalities, provincial capitals, and other cities. Because municipalities and

provincial capitals potentially enjoy special policies from the central and provincial governments,

we use a dummy variable (capital) to capture any benefits. For example, in 1997 Chongqing rose

from an ordinary prefecture-‐level city to become a municipality. After considering its special

history,10 we define Chongqing as a provincial capital in 1990.

Likewise, in the process of Chinese reform and openness, starting in 1980 four special

economic zones (SEZs) were created in China’s coastal areas, and 14 cities later chosen as open

coastal cities (OCCs) in 1994.11 The intention of these SEZs and OCCs was to attract FDI, obtain

access to international markets, and gain special policy advantages. We use the dummies SEZ and

open to measure the effects of this preferential policy. However, as we expect these variables to

correlate highly with distance to the nearest large city and major port, FDI, seaport dummy and

some others, we do not expect significant estimated coefficients. Table 3 provides descriptive

statistics for all the variables.


V. Estimation Results

Table 4 reports the estimated results of the model. In addition to the more conventional

economic growth factors, we include the geographic variables in equations (1)–(3). In equation

(1), we only add the linear item of distance to the nearest large city and major port, while we add

10 Chongqing was the capital of the Republic of China during World War II and is an important city in western China. 11 The four SEZ established in 1980 are Shenzhen, Zhuhai, Xiamen, and Shantou. The 14 OCC in 1994 are Tianjin, Shanghai, Dalian, Qinhuangdao, Yantai, Qingdao, Lianyungang, Nantong, Ningbo, Wenzhou, Fuzhou, Guangzhou, Zhanjiang, and Beihai.

15 / 39

the squares and cubes of distances in equation (2) and drop the insignificant cube of distance to a

large city in equation (3).


5.1. Nonlinearity of the CP Structure in China’s Urban System

In equation (1), both of the distances are insignificantly negative, which means urban

economic growth decreases away from large cities and major ports, but the effect is statistically

insignificant. As discussed, the nonlinearity of the CP model may exist in China’s urban system. If

so, and as Fujita and Mori (1997) and Fujita, Krugman, and Mori (1999) have showed, the

∽-‐shaped correlation between distance and economic activity may be present in our estimated

results. Consequently, we add the squares and cubes of distance in equation (2).

The results of equation (2) show that the distance to a major port and its square and cube

are all significant, which proves nonlinearity in China’s urban system. However, none of the three

terms for distance to a large city is significant. We surmise a ∽-‐shaped curve could only occur

when the distance to a large city is sufficiently long, otherwise, we would only observe a U-‐shape

when using real data. Therefore, we drop the cube of the distance to a large city from equation

(3). Obviously, in equation (3) all distance variables are significant: the distance to the large city

is negative, its squared term is positive. In contrast, the distance to a major port is negative, while

the squared term is positive and the cubed term is negative.

Based on estimated results, we respectively simulate the correlation between the distances

to the major ports or large cities and the urban economic growth rate in Figures 4 and 6. The

horizontal axis represents the distance (kilometers) away from major ports, and the vertical axis

is the urban economic growth rate (%). We discover a CP pattern of urban system evolution

consistent with theory.


The dashed line in Figure 4 suggests that the impact of distance to major port on urban

economic growth has basically the same shape as the market potential curve of the CP model in

the urban system (Fujita and Mori, 1997; Fujita, Krugman, and Mori, 1999). While a city is located

within about 600 km away from a major port, the closer it is to the major port and international

markets, the larger market potential and the higher economic growth rate it has. When the

distance is farther away, international market access is no longer that important. Therefore, a

location far away from a port may promote the accumulation of regional and domestic market

potential, as well as the development of local economies. When distance is sufficiently long

(farther than 1,500 km from a port), cities remote from both the domestic and international

markets suffer from low market potential and a lower economic growth rate. In addition, when

we follow economic geography studies (Dobkins and Ioannides, 2001; Partridge et al., 2009) and

only control for the geographic variables in model (4), the results do not change much and all the

16 / 39

coefficients are statistically more significant. Interestingly, the model including only two distance

variables explains 6.9% of the variance in city growth, compared with 42.8% when all the

variables are included in model (3).

To make the effects of the distances to major ports clearer, we mark the two turning points

of economic growth curve on the map of mainland China in Figure 5. Here, 600 km from Hong

Kong and Shanghai, denoted with a dashed curve, is the location with the lowest average growth,

all other things being equal, while 1,500 km away, marked by a solid curve, is the place with

second-‐highest growth rate.


In Figure 6, we simulate the correlation between the distance to the nearest large city and

urban economic growth using the results of our regressions. When a city is close to a large city,

spatial agglomeration effects promote large cities to absorb economic resources from their

surroundings, which is a significant centripetal force. Meanwhile, being closer to a large city, also

means greater market access for a regional economic center. Therefore, the closer to the central

cities, the faster a city grows. However, when further away from large cities, instead of the

centripetal force, the centrifugal force plays the dominant role. Therefore, the further the

distance, the faster a city grows. Our estimated result shows the turning point is about 300 km,

which means that within about 300 kilometers, the interplay between cities displays a strong

centripetal force, which is similar to the findings in Hanson (2005). The difference is that we also

find that when the distance is greater than 300 km, because of transportation costs and other

external diseconomies, centrifugal force dominates the spatial interaction between cities

performance.


However, in the regional urban systems, there is no sign of the possibility of multicentric

systems. This may be because of the limited population or scope surrounding the cores. In Figure

4, the complete ∽-‐shaped curve requires at least 1,400 km distance, while the real distance to the

nearest large city in regional systems is not sufficiently long. In fact, China’s urban system today

is the result of a long-‐term evolution, thus new large cities may have emerged where there was a

greater market potential because of the spatial agglomeration of established large cities, as

shown by the numerical simulation in Fujita, Krugman, and Mori (1999). Therefore, in Figure 4,

what we can see may be only the left half of the ∽-‐shaped curve, not the whole. Moreover, the

impact of major ports is stronger than large cities on urban economies because within a certain

distance economic growth decreases much faster as distance from major ports increases than it

does away from a large city.

Based on the above, we can safely conclude that the national urban system in China has

presented a complete CP structure because of the adjustment of urban economies to

17 / 39

international markets. The importance of large cities in regional urban systems is less significant

but also verified. We also find evidence for the agglomeration shadow modeled by Krugman

(1993) in the Chinese urban system. This suggests that being closer to an agglomeration center is

not always good news for the local economy. This finding adds new evidence in helping solve the

paradox that Partridge et al. (2009) found between the positive closeness–growth relationship

when using real data and the theoretical hypothesis of an “agglomeration shadow.”

5.2. Border Effect

The same-‐province dummy (samepro) is always significantly negative, which means cities

have a lower growth rate if they are in the same province as their nearest large city. This is

because the absorptive effects of large cities will be larger in this circumstance. This means that

the border effect, similar to the findings in Parsley and Wei (2001) and Poncet (2005) exists

between Chinese provinces, which is likely to increase the actual distances between cities in

different provinces. Based on our estimates, China’s interprovincial border effect is equivalent to

as much as 260 km12 for two neighboring cities. Poncet (2005) finds that the interprovincial

border in China is like an international border in Europe, so our estimate of the interprovincial

border effect is acceptable.

The border effect in this paper presents as the distortion of the spatial concentration. We

believe this is consistent with China’s history of province-‐level market segmentation (Young,

2000; Ponect, 2005) and migration restrictions (Au and Henderson, 2006a, 2006b). Moreover,

Chen et al. (2007) argue that provincial governments have an incentive to restrict the

agglomeration effects from large cities in other provinces using administrative force to protect

their own economic development. Further, Lu and Chen (2009) find that for most observations at

provincial level market segmentation increases economic growth. These conclusions are also

consistent with the findings in this paper. Although in cities of the same province segmentation

can prevent them from the absorptive effects of large cities in other provinces, market

segmentation always brings losses in resource allocative efficiency. In turn, this lowers the

growth rate for the whole economy, resulting in an underdeveloped city scale (Au and

Henderson, 2006a), and a smaller scale inequality among Chinese cities (Fujita et al., 2004).

5.3. Geography and Policy

We also note that in equations (1)–(3), the center, west, and capital dummies are not

significant. This differs from the findings in previous research, including Bao et al. (2002) and Ho

and Li (2010). This is because with the spatial interaction among cities controlled for in our

paper, no other growth disadvantages significantly exist in the western or central regions, and

there are no other obvious advantages in provincial capitals or municipalities. In other words,

spatial agglomeration in China’s urban system contributes most to interregional economic

12 We simply estimate this border effect by dividing the coefficient of the same-‐province dummy by the linear term of the distance to the nearest large city in equation (3).

18 / 39

disparity in China.

Though correlated with the dummy variables for open and SEZ, the dummy seaport is

significantly positive, whereas open and SEZ are insignificant. This implies the possibility that

geography is more important than policy for China’s long-‐run urban growth. Nevertheless, this

needs further study. Furthermore, the riverport dummy is insignificantly positive.

5.4. Other Economic Growth Factors

In our regressions, we also control for other urban economic growth factors, the impacts of

which do not vary significantly across equations (1)–(3). Most of the factors are insignificant,

except for education as measured by the teacher–student ratio. Education (edu) is significantly

positive. This is because education investment in the initial stage promotes human capital

accumulation and economic growth in the long run.

Our regression results also imply that the impact of investment (inve) is insignificantly

negative which may be because Chinese cities with high investment have no obvious economic

advantage in the long term because, as a whole, the Chinese economy suffers from low efficiency

spawned by overinvestment (Zhang, 2003).

Some other factors, like labor (lab), government expenditure (gov), FDI (fdi), urbanization

(urb), population density (density, density2), and industrial structure (SM) all have insignificant

impacts on urban economic growth in the long run based on our regression results. Finally, the

initial level of per capita GDP has an insignificant negative impact on China’s urban economic

growth, showing no significant trend in conditional convergence at city level. The insignificant

variables do not become significant even if we drop the variables for distance to major ports and

large cities. We also exercised estimation by substituting right-‐hand-‐side variables with their

average values across the time span of our data, and the results do not change significantly.13

5.5. Robustness Checks

We estimate several models based on equation (3) to test the robustness of our key findings

in Table 5. In equations (5) and (6), we change the definition of a large city to retest the

hypothesis about regional urban systems. In particular, we redefine large cities as provincial

capitals and municipalities. Thus, distance to the nearest large city is replaced by distance to the

provincial capital (distcap). We again measure the geographic distance using China Map 2008

(Beijing Turing Software Technology Co., Ltd., China Transport Electronic & Audio-‐Video

Publishing House, 2008). We only include the linear term of distance to the provincial capital in

equation (5) and add the squared term to equation (6). For consistency with the variable for

which we control, we drop the dummies samepro and largecity, and replace the variable gdpofbig

by gdpofcap (the economic scale of the provincial capitals, that is, ln(GDP) in 1990 excluding

agricultural output).

In equation (7), we add Tianjin into our definition of major ports and replace distance to the

13 Results available from authors upon request.

19 / 39

nearer major port between Shanghai and Hong Kong (disport) by distance to the nearest major

port among Shanghai, Hong Kong and Tianjin (disport3).

Time span is also an important issue for our study as we highlight the role of China’s

openness and international trade in the changing urban system. However, a large amount of

information in 1992 and 1993 are missing from the Chinese Urban Statistical Yearbooks (National

Bureau of Statistics, 1993–94). Therefore, we use the urban data from 1994 and replace the

dependent variable with the average annual growth rate of real per capita GDP from 1994 to

2006 (dgdp2) deflated by provincial urban CPI, where agricultural output and population are also

excluded. Another benefit of this robustness check is the greater number of observations,

increasing from 133 in equation (3) to 193 in equation (8). Table 5 includes the results of the

robustness checks, with only the variables relevant to spatial interaction reported to save space.


As shown in Table 5, in equation (5), instead of distance to nearest large city and its square,

we only include distance to the provincial capital, which is significantly negative, suggesting the

provincial capitals have strong absorptive effects on the surrounding cities. In equation (6), we

include both the distance to the provincial capital and its square, both of which are not

significant. Consequently, provincial capitals have only significant centripetal forces, which is not

conducive to the economic growth of remote cities. One explanation is that the number of

provincial capitals is about twice that of large cities, so that for most observations the distance to

the provincial capital is much shorter than to the nearest large city. In this context, only a

negative distance–growth relationship would appear, which we can see as the first part of the

U-‐shape we find for the relationship between the distance to the nearest large city and urban

growth. Moreover, the larger initial economic scale of the provincial capital, the slower its urban

economic growth, as gdpofcap is significantly negative. This result shows a stronger absorptive

effect for larger central cities. The variables for the distance to major ports are usually significant.

In equation (7) where Tianjin is included as a major port in China’s urban system, the

variables for distance to major ports are still significant, but the significance levels decrease,

which proves that the economic agglomeration of Tianjin is not as strong as Shanghai and Hong

Kong. We also estimated three separate regressions using the data within each city group. The

∽-‐shaped relationship between distance-‐to-‐port and urban growth is robust in city groups

centered on Shanghai and Hong Kong. However, for the cities centered on Tianjin, the

relationship is insignificant. This is understandable as Tianjin is not comparable with Shanghai

and Hong Kong in terms of economic scale and cargo throughput. What is more, the difference

between Tianjin and other leading ports in the Bohai Belt is not huge in terms of cargo

throughput. If we construct a variable of distance to the four major ports in the Bohai Belt

weighted by their cargo throughputs, then the distance–growth relationship becomes

20 / 39

significant.14 This indicates again that Tianjin is not strong enough to act as a core of a national

level urban system. In addition, the variables for distance to nearest large city are insignificant

with the same sign as in equation (3). This suggests that the robustness of the importance of

large cities and regional urban systems is less than major ports and the national urban system.

In equation (8) with the dependent variable now specified as economic growth during

1994–2006, the significance of the distance-‐to-‐port variables are almost the same as in equation

(3), while the significance of the distance to large cities decreases substantially. It is then possible

that ever since China’s rapid openness to the international market in 1994, the agglomeration

patterns of China’s urban system change to meet the needs of developing international trade, so

that the distance to major ports is becoming more and more important, at least when compared

with that to large cities. However, this is beyond the scope of this paper and further study is

required.

In existing NEG literature, per capita GDP, rather than its growth, usually serves as the

dependent variable. In our study, we also attempt to substitute the growth rate of per capita GDP

with its level, and conduct panel data regression and year-‐by-‐year estimation. To ensure that our

results are robust, even if we only consider the manufacturing and service sector, we regress GDP

per worker for secondary and tertiary industries on the geographic and other explanatory

variables. These additional regression exercises do not alter our major findings significantly.15

VI. How Important Is Geography? Long-‐Run, Medium-‐Run and Short-‐Run Comparison

There are three possible explanations for the insignificance of the growth factors in the long

run. First, the most obvious is that these factors may have no long-‐term impact on urban

economic growth. Second, measurement error in the variables and small sample size may have

reduced their levels of significance. To expand the sample size, we extend our long-‐run growth

regression to medium-‐ and short-‐run analyses.

In Table 6, there are three equations for urban economic growth in the short, medium and

long run, respectively. In equation 9, to minimize the simultaneity bias, we lag the explanatory

variables by 1 year and use the city-‐level panel data during 1990-‐2006 to run the regression for

yearly growth. In equation 10, per capita compound average GDP growth rate over 5-‐year

periods, namely, 1990-‐1995, 1996-‐2000, and 2001-‐2005, is used as dependent variable. The

explanatory variables are initial values over those 5-‐year periods. Again, random-‐effects GLS

estimations are employed, since geographic variables are time-‐invariant. Equation 3 is duplicated

for easy comparison with short-‐ and medium-‐run growth regressions. In all the three equations,

both distance-‐to-‐ports and distance-‐to-‐large cities are significant, showing the importance of

geography through short to long run. In equation 3 duplicated from Table 4, the sample size is the

smallest, but the distance-‐to-‐ports variables have the most significant effects. So in the long run,

international market is more important.

14 These results are not reported to conserve space, but available upon request. 15 We do not report these results to save space, but they are available upon request.

21 / 39


Based on the estimation results, we add to Figure 4 a solid line simulating the relationship

between distance-‐to-‐ports and growth in the short run. The two curves lie close to each other,

showing that the long-‐run relationship between distance and growth is the accumulative effect of

that in the short run.

It’s quite interesting to compare other variables’ effects on growth in the short, medium and

long run. Investment-‐to-‐GDP ratio has positive effect on growth in the short run, but the

magnitude of the coefficient shrinks in the medium run, and it becomes insignificant and negative

in the long run. This is alarming for China’s growth pattern that highly relies on investment.

Although in the short run investment enhances growth, in the long run, more investment does

not generate higher growth. Inefficiency and poor returns to investment might be the potential

reason (Zhang, 2003). More recent data analysis shows that Middle and Western China has seen a

faster capital deepening process during 1997-‐2007, which worsens the efficiency and returns to

investment (Brandt, Tombe and Zhu, 2010; Lu, 2013). Similarly, government intervention

significantly increases growth in the short run, since it may boost demand. But it has no

significant effects in the long run, because government intervention distorts the market and

offsets its growth-‐enhancing effects. 16 In the medium-‐ and long-‐run regressions, we substitute

the initial value of those time-‐variant variables by their average values of over 5 or 16 years, the

insignificant coefficients don’t change into significant. So, the insignificant effect is not because

the short-‐run positive effect fades away as time goes by.17

Labor has negative effects on growth in the short run, as in the period we study China still

had surplus labor. And labor is not significant in the long-‐run. In contrast, education significantly

increases growth, and the effect is even greater and more statistically significant in the long run.

The education measurement, teacher-‐student ratio, is actually an investment variable that has

lagged effects on growth, and this may partly explain why education is more important in the

long run. According to our estimation, if the number of teacher is increased by 1 per 100 students,

the long-‐run growth would be 1.07 percentage points higher. FDI does not affect growth, since

FDI correlates with geography. After controlling for a lot of geography variables, FDI does not

have separable growth-‐enhancing effects. Finally, cities with lower initial per capita income level

grow faster, showing a conditional convergence in the short run, but this convergence effect does

not exist in the long run.

For a comparison, we drop the distance variables and the dummy for whether the nearest

big city is in the same province, and run traditional growth regressions. In the long-‐run model,

the labor and investment variables become significant at 10% level, and the provincial capital

dummy turns significant at 5% level. In other words, some traditional growth factors are

16 Usually, in the long-‐run growth regression, government expenditure-‐to-‐GDP ratio has negative effects (Barro, 2000)。 17 This result is available upon request.

22 / 39

correlated with the geographic variables we construct for the locations of cities. So, geography is

a more fundamental factor to explain urban economic growth. 18

Furthermore, we want to know how important geography is relative to other variables. Can

other growth-‐enhancing factors alleviate geographic disadvantages? This is relevant for regional

development policy making. To answer this question, we only control variables other than the

distance-‐to-‐ports, distance-‐to-‐large cities variables and their non-‐linear terms. The results are

reported in Table 7. In the short-‐run growth model, 34.6% of variations in growth are explained

by model (9), and in model (11) the two distance variables explain 0.2% of growth variations.

Another comparison is between model (3) and (4) for long-‐run growth. In model (3), all the

explanatory variables explain 40.4% of growth, while the two distances explain 6.9% in model

(4). This is another signal that geography plays an essential role in long-‐run economic growth at

city level, and much more powerful than its short-‐run role.


To compare the effects of distances with and without other variables more visible, we

simulate the distance-‐growth relationship in Figure 7. It shows that without controlling for other

variables, the effect of distance-‐to-‐ports is much smaller than that with other factors controlled

for. The difference is greater in middle China than the remote interior areas, so there must be

something that can help counteract the geographic disadvantages. Based on equation (3),

education investment is the major factor that significantly enhances long-‐run growth. In China,

the middle regions have even slightly higher educational level, measured by per capita schooling

years, than the eastern cities do (Lu, 2013). So, human capital investment is a policy tool that can

be used to enhance growth in the geographically disadvantageous regions. In contrast, as shown

already, moving money by investment and government expenditure is not effective to help

laggard regions in the long run.


VII. Conclusions

This paper represents an early attempt to verify the spatial pattern of China’s urban system

following the non-‐linear CP model. China’s large territory and its open policy since the 1990s

makes China a feasible application and suitable experimental field for testing the non-‐linear CP

model of urban systems.

The most important finding of this paper is to verify the ∽-‐shaped nonlinear correlation

between the geographical distance to major ports and urban economic growth, which is

consistent with the non-‐linear CP model of urban systems. We find that China’s urban economies

adjust to increasingly important international trade according to their access to global market,

especially after China’s drastic opening up since the mid-‐1990s. Compared with distance to port, 18 This result is available upon request.

23 / 39

the distance to the nearest large city has a U-‐shaped relationship with urban growth.

The agglomeration shadow first modeled by Krugman (1993) is also evident in China’s

urban system, which means that being closer to agglomeration centers is not always good news

for the local economy. We find that within a scope about 600 km away from Shanghai and Hong

Kong growth rate falls with distance to major ports, while growth goes up for regions 600-‐1200

km away and falls again for more remote areas. This finding adds new evidence in helping solve

the paradox that Partridge et al. (2009) found between the positive closeness–growth

relationship found when using real data and the theoretical hypothesis of an agglomeration

shadow.

Our study also shows that in the long run, geography plays a much more important role than

in the short run, in terms of the significance levels of the distance-‐to-‐ports variables, their

magnitudes and explanatory power in the model. Investment and government expenditure can

increase short-‐run growth, but not significantly in the long run. Only educational investment is

found effective to help alleviate geographic disadvantage for laggard areas in the long run.

Therefore, regional policies for balanced development should be “moving people”, in other words,

free mobility of labor out of disadvantageous regions, rather than “moving money” by physical

capital investment and government-‐sponsored projects. The only effective “moving money”

policy is to invest in human capital in laggard regions to make people more capable of making

choices between moving out and staying.

24 / 39

References:

1. Anderson, G., Ying Ge, 2005. The Size Distribution of Chinese Cities, Regional Science and

Urban Economics 35, 756–776.

2. Au, Chun-‐Chung, J. Vernon Henderson, 2006a. Are Chinese Cities Too Small? Review of

Economic Studies 73, 549–576.

3. Au, Chun-‐Chung, J. Vernon Henderson, 2006b. How Migration Restrictions Limit

Agglomeration and Productivity in China, Journal of Development Economics 80,

350–388.

4. Baldwin, R. E., R. Forslid, P. Martin, G.I.P. Ottaviano and F. Robert-‐Nicoud, 2003,

Economic Geography and Public Policy, Princeton: Princeton University Press.

5. Bao, Shuming, Gene Hsin Chang, Jeffrey D. Sachs, Wing Thye Woo, 2002. Geographic

Factors and China’s Regional Development under Market Reforms, 1978–1998, China

Economic Review 13, 89–111.

6. Barro, Robert J., 2000. Inequality and Growth in a Panel of Countries, Journal of

Economic Growth 5, 87–120.

7. Black, Duncan, J. Vernon Henderson, 1999. Spatial Evolution of Population and Industry

in the United States, American Economic Review 89, 321–327.

8. Brandt, Loren, Trevor Tombe and Xiaodong Zhu, 2010. Factor Market Distortions across

Time, Space, and Sectors in China, University of Toronto Working Paper.

9. Brülhart, M., P. Koenig, 2006. New Economic Geography Meets Comecon, Economics of

Transition 14, 245–267.

10. Chen, Min, Qihan Gui, Ming Lu, Zhao Chen, 2007. Economic Opening and Domestic

Market Integration, in Ross Garnaut and Ligang Song (eds.), China: Linking Markets for

Growth, Asia Pacific Press, 369–393.

11. Chen, Zhao, Yu Jin, Ming Lu, 2008. Economic Opening and Industrial Agglomeration in

China, in M. Fujita, S. Kumagai and K. Nishikimi (eds.), Economic Integration in East Asia,

Perspectives from Spatial and Neoclassical Economics, Edward Elgar Publishing,

276–315.

12. Clarke, G. R. G., 1995, “More Evidence on Income Distribution and Growth,” Journal of

Development Economics, 47, 403–427.

13. Combes, Pierre-‐Philippe, 2000. Economic Structure and Local Growth: France,

1984–1993, Journal of Urban Economics 47, 329–355.

14. Dobkins, L., Y. Ioannides, 2000. Dynamic Evolution of the Size Distribution of U.S. Cities,

in J. M. Huriot, J. F. Thisse (eds.), Economics of Cities, Cambridge: Cambridge University

Press, 217–260.

15. Dobkins, L., Y. Ioannides, 2001. Spatial Interactions among U.S. Cities: 1900–1990,

Regional Science and Urban Economics 31, 701–731.

16. Fujita, M., 1988. A Monopolistic Competition Model of Spatial Agglomeration, Regional

Science and Urban Economics 18, 87–124.

25 / 39

17. Fujita, M., V. Henderson, Y. Kanemoto, T. Mori, 2004. Spatial Distribution of Economic

Activities in Japan and China, in V. Henderson and J. F. Thisse (eds.), Handbook of Urban

and Regional Economics, vol.4, North-‐Holland, 2911–2977.

18. Fujita, M., P. R. Krugman, 2004. The New Economic Geography: Past, Present and the

Future, Regional Science 83, 139–164.

19. Fujita, M., P. R. Krugman, and T. Mori, 1999. On the Evolution of Hierarchical Urban

Systems, European Economic Review 43, 209–251.

20. Fujita, M., P.R. Krugman, and A.J. Venables, 1999. The Spatial Economy: Cities, Regions

and International Trade, Cambridge, Massachusetts: MIT Press.

21. Fujita, M., T. Mori, 1996. The Role of Ports in the Making of Major Cities:

Selfagglomeration and Hub-‐effect, Journal of Development Economics 49, 93–120.

22. Fujita, M., T. Mori, 1997. Structural Stability and Evolution of Urban System, Regional

Science and Urban Economics 27, 399–442.

23. Fujita, M., J. F. Thisse, 1996. Economics of Agglomeration, Journal of the Japanese and

International Economies 10, 339–378.

24. Glaeser, E., J. Scheinkman, and A. Shleifer, 1995. Economic Growth in a Cross-‐section of

Cities, Journal of Monetary Economics 36, 117–143.

25. Hanson, Gordon H., 1998. Regional Adjustment to Trade Liberalization, Regional Science

and Urban Economics 28, 419–444.

26. Hanson, Gordon H., 2001. Scale Economies and Geographic Concentration of Industry,

Journal of Economic Geography 1, 255–276.

27. Hanson, Gordon H., 2005. Market Potential, Increasing Returns, and Geographic

Concentration, Journal of International Economics 67, 1–24.

28. Ho, Chunyu, Dan Li, 2010. Spatial Dependence and Divergence across Chinese Cities,

Review of Development Economics 14, 386–403.

29. Ioannides, Y. M., H. G. Overman, 2004. Spatial Evolution of the US Urban System, Journal

of Economic Geography 4, 131–156.

30. Krugman, P., 1991. Increasing Returns and Economic Geography, Journal of Political

Economy 99, 483–499.

31. Krugman, P., 1993. First Nature, Second Nature, and Metropolitan Location, Journal of

Regional Science 33, 129–144.

32. Krugman, P., 1996. Confronting the Mystery of Urban Hierarchy, Journal of the Japanese

and International Economies 10, 399–418.

33. Krugman, P., R. L. Elizondo, 1996. Trade Policy and the Third World Metropolis, Journal

of Development Economics 49, 137–150.

34. Lu, Ming, Zhao Chen, 2009. Fragmented Growth: Why Economic Opening May Worsen

Domestic Market Segmentation? (in Chinese), Economic Research Journal 3, 42–52.

35. Lu, Ming, 2013, The Power of Space: Geography, Politics and Urban Development, (in

Chinese), Shanghai People’s Press.

26 / 39

36. National Bureau of Statistics, Chinese Cities Statistical Yearbook (1991–2007), China

Statistics Press, Beijing, 2007.

37. Overman, Henry G., Y. M. Ioannides, 2001. Cross-‐Sectional Evolution of the U.S. City Size

Distribution, Journal of Urban Economics 49, 543–566.

38. Parsley, D., S. Wei, 2001. Explaining the Border Effect: The Role of Exchange Rate

Variability, Shipping Cost, and Geography, Journal of International Economics 55,

87–105.

39. Partridge, M. D., D. S. Rickman, K. Ali and M. R. Olfert, 2009. Do New Economic Geography

Agglomeration Shadows Underlie Current Population Dynamics across the Urban

Hierarchy? Papers in Regional Science 88, 445–466.

40. Poncet, S., 2005. A Fragmented China: Measure and Determinants of Chinese Domestic

Market Disintegration, Review of International Economics 13, 409–430.

41. Venables, A. J., 1996. Equilibrium Locations of Vertically Linked Industries, International

Economic Review, 37, 341–359.

42. Wan, Guanghua, Ming Lu, Zhao Chen, 2007. Globalization and Regional Income Inequality:

Empirical Evidence from within China, Review of Income and Wealth 53, 35–59.

43. Wei, S., 1995. The Open Door Policy and China’s Rapid Growth: Evidence from City-‐level

Data, in Takatoshi Ito, Anne O. Krueger (eds.), Growth Theories in Light of the East Asian

Experience, NBER-‐East Asian Seminar on Economics, Vol. 4, University of Chicago Press.

44. Young, A., 2000. The Razor’s Edge: Distortions and Incremental Reform in the People’s

Republic of China, Quarterly Journal of Economics 115, 1091–1135.

45. Zhang, Jun, 2003. Investment, Investment Efficiency, and Economic Growth in China,

Journal of Asian Economics 14, 713–734

27 / 39

Figure 1: Share of Industry Value-‐Added in GDP (%) in China, US and France (1980–2008)

Sources: World Bank National Accounts data and OECD National Accounts data files.

Figure 2: International Trade, Urbanization, and GDP Growth in China (1978–2008)

Source: NBS, China Statistical Abstract 2009.

28 / 39

HeilongJiang

Liaoning

Jinlin

Beijing [Inner Mongolia]

Tianjin Shandong

Jiangsu

Shanghai

Zhejiang

Fujian

g Guangdong [Guangxi]

[Tibet]

[Xinjiang]

[Ningxia]

[Sichuan]

[Qinhai]

[Gansu]

[Guizhou] [Yunnan]

[Chongqing]

[Shaanxi]

Hainan

（Hebei）

（Henan）

（Shanxi）

（Anhui）

（Jiangxi）（Hunan）

（Hubei）

Tianjin

Hong Kong

Figure 3: Distribution of Large Cities, Major Ports, and Provinces in Mainland China

Note: (1) Deltas and squares are large cities and major ports using our definitions, respectively.

(2) Provinces in square brackets are located in the West, those in parentheses are in the

Center, and the remainders are in the East.

29 / 39

-9

-8

-7

-6

-5

-4

-3

-2

-1

00 200 400 600 800 1000 1200 1400 1600 1800 2000

Distance (km)

Gro

wth

(%)

Long runShort run

Figure 4: Distance to Major Port and Urban Economic Growth

30 / 39

Figure 5: Geographic Distance and Urban Economic Growth

-1.5

-1.3

-1.1

-0.9

-0.7

-0.5

-0.3

-0.1 0 50 100 150 200 250 300 350 400 450

Distance (km)

Gro

wth

(%)

Figure 6: Distance to Large City and Urban Economic Growth

31 / 39

-6

-5

-4

-3

-2

-1

00 200 400 600 800 1000 1200 1400 1600 1800 2000

Distance (km)

Gro

wth

(%)

Other variables controlledOnly distance variables controlled

Figure 7: Distance to Major Ports and Urban Economic Growth

32 / 39

Table 1: Number of China’s Cities

Year Total Municipalities Prefecture-‐level cities County-‐level cities

1990 467 3 185 279 1995 640 3 210 427 2000 663 4 259 400 2005 657 4 283 370

Source: NBS, Chinese Urban Statistical Yearbooks (1991–2006).

Note：1）Our database does not include information for Hong Kong, Macao, Taiwan and Tibet.

2) Sub-‐provincial cities are also included in the prefecture-‐level cities.

Table 2: China’s Urban Hierarchy

Year Prefecture-‐ level and above

cities

Provincial capitals

Nonagricultural population>2

million

Nonagricultural population>1.5

million

Nonagricultural population>1

million 1990 185 29 9 14 31 1995 210 29 10 16 32 2000 259 30 13 18 38 2005 286 30 22 31 54

Source: NBS, Chinese Urban Statistical Yearbooks (1991–2006).

Note：Provincial capitals include both municipalities and other provincial capitals.

33 / 39

Table 3: Descriptive Statistics

Variable Obs Mean Std. Dev. Min. Max. dgdp(%) 208 8.36 2.82 –0.37 17.10 distbig(km) 286 291.33 251.35 0 2351.8 disport(km) 286 896.75 542.90 0 3526.4 largecity 286 0.07 0.26 0 1

gdpofbig(100 million yuan)

286 114.93 80.74 37.36 344.47

samepro 286 0.41 0.49 0 1 seaport 286 0.11 0.32 0 1 riverport 286 0.08 0.28 0 1 center 286 0.35 0.48 0 1 west 286 0.29 0.46 0 1

gdp per capita(yuan) 210 3581.30 2120.26 382.23 19820.83 inve(%) 210 25.91 16.67 1.54 127.88 labor(%) 209 57.60 7.79 30.01 93.79 edu(%) 205 4.61 0.803 3.02 7.48 gov(%) 199 10.27 3.82 1.04 29.16 fdi(%) 145 3.28 9.78 0.01 100.99 urb(%) 211 59.40 23.62 8.05 96.53

density (per km2) 210 11849.77 4180.96 3631 32920 SM(%) 210 59.53 31.95 7.14 188.51 capital 286 0.11 0.31 0 1 open 286 0.05 0.21 0 1 SEZ 286 0.02 0.15 0 1

34 / 39

Table 4: Geography and Urban Economic Growth

(Dependent variable is the compound average growth rate between 1990 and 2006.)

(1) (2) (3) (4) dgdp dgdp dgdp dgdp disport/103 –0.215 –12.9** –11.2** –7.37*** (0.716) (5.55) (4.29) (2.79) disport2/106 15.0** 12.8*** 8.81*** (6.79) (4.82) (3.03) disport3/109 –4.98** –4.17** –3.14*** (2.36) (1.60) (0.995) distbig/103 –1.69 –4.99 –8.02** –5.58** (1.49) (7.35) (3.52) (2.21) distbig2/106 6.76 13.2** 12.1*** (14.8) (5.40) (3.54) distbig3/109 3.57 (7.59) largecity 0.754 0.883 0.724 (0.918) (0.988) (0.925) gdpofbig0 –0.549 –0.742 –0.763 (0.625) (0.637) (0.633) samepro –1.947*** –2.165*** –2.211*** (0.666) (0.707) (0.697) seaport 1.978** 2.022** 1.962** (0.838) (0.843) (0.831) riverport 0.752 0.730 0.614 (0.649) (0.693) (0.645) center –1.391* –0.989 –1.025 (0.797) (0.812) (0.806) west –0.980 –1.349 –1.330 (0.878) (0.878) (0.874) lngdp –0.845 –1.249* –1.201 (0.726) (0.745) (0.735) inve/102 –3.45* –2.78 –2.74 (2.01) (2.01) (2.00) labor/102 4.27 4.16 4.11 (3.36) (3.44) (3.43) edu 0.552*** 0.590*** 0.584*** (0.172) (0.171) (0.170) gov/102 4.39 2.69 3.07 (7.43) (7.53) (7.46) fdi/102 3.05 1.46 1.32 (2.56) (2.62) (2.59) urb/102 0.70 0.35 0.46 (1.35) (1.36) (1.33) density/104 2.15 2.70 2.80 (2.07) (2.08) (2.06) density2/109 –5.07 –7.77 –7.99 (6.00) (6.05) (6.01) SM/103 –8.87 –5.16 –4.75 (10.6) (10.7) (10.6) capital 1.051 0.598 0.548 (0.775) (0.792) (0.782) open –0.348 –0.287 –0.282 (1.043) (1.031) (1.027) SEZ –1.798 –1.976 –2.005 (1.436) (1.416) (1.410) Constant 13.06* 19.72** 19.39** 10.39*** (7.626) (8.290) (8.230) (0.847)

35 / 39

Observations 132 132 132 202 R2 0.387 0.430 0.428 0.069

Notes: (1) Standard errors in parentheses;

(2) * p < 0.10, ** p < 0.05, *** p < 0.01.

36 / 39

Table 5: Robustness: Distance and Urban Economic Growth

(3) (5) (6) (7) (8) dgdp dgdp dgdp dgdp dgdp2 disport/103 –11.2** –3.73 –3.84 –13.9*** (4.29) (2.50) (2.53) (5.27) disport2/106 12.8*** 3.59* 3.70* 14.6** (4.82) (2.10) (2.13) (6.19) disport3/109 –4.17** –0.854* –0.876* –4.28** (1.60) (0.463) (0.4.69) (2.07) disport3/103 –8.81 (5.54) disport32/106 14.4* (7.55) disport33/109 –7.42** (3.30) distbig/103 –8.02** –9.53** –9.25* (3.52) (4.30) (5.40) distbig2/106 13.2** 14.7** 10.7 (5.40) (5.68) (8.04) distcap/103 –7.48*** –10.4 (2.50) (8.53) distcap2/106 7.05 (19.8) gdpofcap –2.011*** –2.036*** (0.423) (0.430) Other controls Y Y Y Y Y Observations 132 132 132 132 192 R2 0.428 0.492 0.493 0.434 0.373 Notes: (1) Standard errors in parentheses;

(2) * p < 0.10, ** p < 0.05, *** p < 0.01.

37 / 39

Table 6: Urban Economic Growth in the Short, Medium and Long Run

(9) (10) (3) short run medium run long run

disport/103 -‐15.6** -‐16.3*** –11.2** (6.8) (5.8) (4.29) disport2/106 18.0** 17.4*** 12.8*** (7.5) (6.5) (4.82) disport3/109 -‐6.09** -‐5.41** –4.17** (2.40) (2.12) (1.60) distbig/103 -‐17.2*** -‐10.8** –8.02** (5.5) (4.5) (3.52) distbig2/106 22.5*** 16.3** 13.2** (8.1) (7.1) (5.40) largecity 1.436 2.354* 0.724 (1.849) (1.424) (0.925) gdpofbig0 -‐1.355 -‐0.301 –0.763 (1.083) (0.886) (0.633) samepro -‐5.359*** -‐4.171*** –2.211*** (1.112) (0.907) (0.697) seaport 3.298** 3.198** 1.962** (1.588) (1.247) (0.831) riverport -‐0.111 0.548 0.614 (1.309) (1.012) (0.645) center -‐0.138 -‐0.287 –1.025 (1.358) (1.102) (0.806) west -‐0.365 -‐0.004 –1.330

(1.485) (1.221) (0.874) lngdp -‐2.680*** -‐2.858*** –1.201 (0.809) (0.720) (0.735) inve/102 8.7*** 3.5** –2.74 (2.0) (1.7) (2.00) labor/102 -‐10.4*** -‐10.0*** 4.11 (2.0) (1.6) (3.43) edu 0.482 0.289 0.584*** (0.468) (0.321) (0.170) gov/102 26.9*** 16.7** 3.07 (8.9) (7.5) (7.46) fdi/102 1.6 2.4 1.32 (18.9) (17.5) (2.59) urb/102 6.7*** 1.5 4.60 (1.9) (1.6) (13.3) density/104 -‐17.50*** -‐0.77 2.80 (1.50) (3.04) (2.06) density2/109 68.20*** 8.48 –7.99 (2.25) (11.00) (6.01) SM/103 -‐5.72 -‐7.19 –4.75 (8.52) (7.22) (10.6) capital 0.549 -‐0.248 0.548 (1.558) (1.186) (0.782) open -‐0.614 -‐0.050 –0.282 (2.070) (1.587) (1.027) SEZ -‐5.811** -‐2.364 –2.005 (2.801) (2.216) (1.410) Constant 49.38*** 41.05*** 19.39**

38 / 39

(11.10) (9.52) (8.23) Observations 2793 561 132 R2 0.346 0.170 0.428


(2) * p < 0.10, ** p < 0.05, *** p < 0.01.

39 / 39

Table 7：Urban Economic Growth in the Short, Medium and Long Run (Only Distance Controlled)

(11) (12) (4) short run medium run long run

disport/103 -‐11.8** -‐2.7 –7.37*** (5.9) (4.0) (2.79) disport2/106 13.1** 4.5 8.81*** (6.2) (4.2) (3.03) disport3/109 -‐4.04** -‐1.88 –3.14***

(1.99) (1.37) (0.995) distbig/103 -‐7.1 -‐3.9 –5.58** (5.0) (3.2) (2.21) distbig2/106 12.8* 8.7* 12.1*** (7.3) (4.9) (3.54) Constant 12.85*** 8.82*** 10.39***

(1.92) (1.24) (0.847) Observations 3185 706 202 R2 0.002 0.006 0.069


(2) * p < 0.10, ** p < 0.05, *** p < 0.01.

agglomeration$shadow:$!4!/39!!...

Documents